One of the most difficult tasks among social intelligence tasks that read the intentions of others is to estimate their own state from the actions of others. This behavior is one of the characteristics of human intelligence. For example, in the field of psychology, self-image that can not be seen by myself is called blind self and self-recognition of blind self is a measure of growth.
In this research, we used a cooperative game called Hanabi as an artificial intelligence task competing for such blind self. Hanabi is a cooperative type card game where all the agents cooperate to gather scores. All players cooperate with fireworks of five colors represented by rows of cards 1 to 5 to build up. And the size of this fireworks is the score. In this game, the player can not see his card, but can know all the agents’ cards other than his / her own card. Also, in this game communication between agents is restricted. Each agent must consume resources for information transmission in order to teach the number or color of cards of other players. No other communication means are prepared. I implemented an artificial intelligence agent to solve this Hanabi did. This agent can simulate the viewpoints and actions of others. By doing this, we examined how the reproduction of the viewpoint of others leads to scoring.
- Hirotaka Osawa. 2015. Solving Hanabi : Estimating Hands by Opponent’s Actions in Cooperative Game with Incomplete Information. In AAAI workshop: Computer Poker and Imperfect Information, 37–43.
- 大澤博隆. 2015. 協力ゲームHanabiにおけるエージェント間の協調行動の分析 Estimation of Own State by Opponent’s Behavior in Cooperative Game Hanabi. 人工知能学会全国大会論文集 29: 1–4. Retrieved from http://ci.nii.ac.jp/naid/40020492009/en/